Best-Practices for Large Eddy Simulations (LES) in

FLUENT

General Tips


1. Mesh
a. Hex mesh is an ideal choice for accuracy
b. Avoid abrupt change in grid spacing (cell size) or too aggressive
stretching in eddy-congested regions (e.g., boundary layer, near-
wake)
c. Take advantage of hybrid mesh and local mesh refinement
capability (e.g. wall boundary layer, near-wake)
2. Spatial discretization
a. The “node-based gradient” option is preferred for unstructured
(e.g., tet) meshes
b. Stick with central differencing or bounded central differencing for
momentum equations
c. When undershoot/overshoot of the solution fields becomes a critical
issue, consider using high-order upwind schemes (QUICK,
MUSCL, SOU) for scalars
3. Time-discretization
a. Use the second-order scheme (default)
b. Use the NITA/fractional-step method for incompressible flows
c. With the iterative scheme, SIMPLEC has been found best choice
i. When the mesh quality is not bad, jack up the URF (0.9 – 1.0) for
pressure and momentum equations
4. SGS modeling
a. Dynamic Smagorinsky’s model is recommended
b. Dynamic TKE model can potentially benefit highly non-equilibrium
flows and reacting flows


Recommended Procedure

1. Compute the mean flow with steady RANS
k-ε model is sufficient
The RANS solution doesn’t have to be fully converged
2. Superimpose the synthesized turbulence on the mean flow
Use TUI (“/solve/initialize/init-instantaneous-vel”)
3. Switch to LES, select the SGS turbulence model of your
choice
GUI (Define/Model/Viscous…)
4. Select the solver algorithm (e.g., ITA/NITA,
FSM/PISO/SIMPLEC) and the discretization schemes
 Define/Model/Solver…
Solve/Controls/Solution…
5. Set the time-step (∆t) and adjust the solver parameters if
needed (e.g., URF’s, convergence criteria)
Solve/Iterate…
Solve/Controls/Solution…
6. Set the monitors for relevant global (e.g., forces/moments)
and local quantities (e.g., velocity, pressure) of your choice
Solve/Monitors/Force…
Solve/Monitor/Surface…
Solve/Monitor/Volume…
7. Set the autosave of the data files (e.g., every a few hundreds
time-steps)
File/Write/Autosave…
8. Start the transient run and continue until a statistically
stationary state is reached
Monitoring integral and/or local quantities will help you judge this
9. Save the viewgraphs of your choice for animation (contours
of pressure, iso-surfaces of vorticity, second-invariant, etc.)
Solve/Execute Commands…
10. Start sampling the data (to compute mean and r.m.s. values)
Solve/Initialize/Reset Statistics
Solve/Iterate (Click on the “Data Sampling for Time Statistics”
button)
11. Continue sampling for a sufficiently long period of time
Several flow-through times (L/U0)
Until the mean fields recover any homogeneity (e.g., axisymmetry,
2-D)
12. Post-process the results
Display/Contours…(Vectors…)/Unsteady Statistics
Plot/XY Plot…
Plot/FFT…